mcp-horizon-support-v1

Fine-tuned Qwen3-4B on MCP Horizon product documentation for question-answering and support use cases.

Base model

Qwen/Qwen3-4B — fine-tuned with QLoRA (4-bit) via Unsloth.

Training

Setting	Value
Method	QLoRA (LoRA rank 16, alpha 32)
Epochs	5
Training examples	534
Final loss	0.7498
Hardware	NVIDIA RTX 4070 12GB

Evaluation (DeepEval, 60 examples)

Metric	Score	Pass Rate
Answer Relevancy	0.86	80%
Correctness (GEval)	0.32	22%
Completeness (GEval)	0.35	25%

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "mcphorizon/mcp-horizon-support-v1"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype=torch.bfloat16, device_map="auto"
)

messages = [{"role": "user", "content": "How do I configure access controls in MCP Horizon?"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Ollama

A GGUF Q4_K_M version is also available for use with Ollama/llama.cpp.